* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download PDF
Heaven and Earth (book) wikipedia , lookup
ExxonMobil climate change controversy wikipedia , lookup
Michael E. Mann wikipedia , lookup
Soon and Baliunas controversy wikipedia , lookup
Numerical weather prediction wikipedia , lookup
Politics of global warming wikipedia , lookup
Fred Singer wikipedia , lookup
Global warming wikipedia , lookup
Climate resilience wikipedia , lookup
Climatic Research Unit email controversy wikipedia , lookup
Climate change denial wikipedia , lookup
Climate change feedback wikipedia , lookup
Climate engineering wikipedia , lookup
Climate change adaptation wikipedia , lookup
Atmospheric model wikipedia , lookup
Instrumental temperature record wikipedia , lookup
Economics of global warming wikipedia , lookup
Effects of global warming on human health wikipedia , lookup
Climate governance wikipedia , lookup
Carbon Pollution Reduction Scheme wikipedia , lookup
Climate change in Tuvalu wikipedia , lookup
Citizens' Climate Lobby wikipedia , lookup
Solar radiation management wikipedia , lookup
Climate sensitivity wikipedia , lookup
Attribution of recent climate change wikipedia , lookup
Effects of global warming wikipedia , lookup
Media coverage of global warming wikipedia , lookup
Climatic Research Unit documents wikipedia , lookup
Global Energy and Water Cycle Experiment wikipedia , lookup
Public opinion on global warming wikipedia , lookup
Scientific opinion on climate change wikipedia , lookup
Climate change in the United States wikipedia , lookup
Climate change in Saskatchewan wikipedia , lookup
Climate change and poverty wikipedia , lookup
Effects of global warming on humans wikipedia , lookup
Surveys of scientists' views on climate change wikipedia , lookup
General circulation model wikipedia , lookup
IPCC Fourth Assessment Report wikipedia , lookup
Journal of Agricultural and Applied Economics, 45,4(November 2013):719–737 Ó 2013 Southern Agricultural Economics Association Using a Climate Index to Measure Crop Yield Response Ruohong Cai, Jeffrey D. Mullen, John C. Bergstrom, W. Donald Shurley, and Michael E. Wetzstein Using principal component analysis, a climate index is developed to estimate the linkage between climate and crop yields. The indices based on three climate projections are then applied to forecast future crop yield responses. We identify spatial heterogeneity of crop yield responses to future climate change across a number of U.S. northern and southern states. The results indicate that future hotter/drier weather conditions will likely have significant negative impacts on southern states, whereas only mild impacts are expected in most northern states. Key Words: climate change, crop yields, principal component analysis JEL Classifications: Q1, Q54 Contemporary Global Climate Models (GCMs), including the Australian CSIRO 3.5, Canadian CGCM 3.1, and Japanese MIROC 3.2, all predict that average temperature will keep rising with modest changes in precipitation for most states in the continental United States for the rest of the century (Coulson et al., 2010). This is assuming that greenhouse gas emissions Ruohong Cai is a postdoctoral researcher associate, Program in Science, Technology and Environmental Policy, Woodrow Wilson School of Public and International Affairs, Princeton University, Princeton, New Jersey. Jeffrey D. Mullen is an associate professor, Department of Agricultural and Applied Economics, University of Georgia, Athens, Georgia. John C. Bergstrom, W. Donald Shurley, and Michael E. Wetzstein are professors, Department of Agricultural and Applied Economics, University of Georgia, Athens, Georgia. We are grateful to the Editor, Darrell Bosch, and three anonymous reviewers for their helpful comments and suggestions. All remaining errors are the authors’. follow the IPCC SRA1B scenario.1 Although agricultural technologies continue to improve, previous studies have indicated that temperature and precipitation variations have significant impacts on crop yields (Lobell, Cahill, and Field, 2007; Almaraz et al., 2008; Schlenker and Roberts, 2009). Environmental conditions such as soil properties are expected to result in spatially varying climate change impacts. To date, there are a limited number of studies that have attempted to compare the effects of climate variations on crop yields across regions. Tao et al. (2006) 1 The IPCC SRA1B scenario represents a future world of very rapid economic growth, low population growth, and rapid introduction of new and more efficient technology. Major underlying themes are economic and cultural convergence and capacity building with a substantial reduction in regional differences in per capita income. In this world, people pursue personal wealth rather than environmental quality (IPCC, 2007). 720 Journal of Agricultural and Applied Economics, November 2013 studied data from sample stations located in various geographic and climatic zones in China and found that temperature was negatively correlated with crop yield at all stations except Harbin in northeastern China. McCarl, Villavicencio, and Wu (2008) found that the effects of temperature on crop yields vary across U.S. regions. In the appendix of Schlenker and Roberts (2009), the United States was divided into three regions: the northern, the interior, and the southern to explore how the temperature– yield relationship varies over different regions. They found that the threshold where temperature negatively affects yield is slightly lower in warmer areas, and the southern region has a lower sensitivity to extreme heat. Using a crop growth model, Butterworth et al. (2009) found that climate change would increase the productivity of oilseed rape in the United Kingdom but with the greatest benefits in Scotland in the north rather than England in the south. Understanding how crop yield responses vary across regions can help predict the price and welfare impacts of climate change and aid in planning mitigation strategies related to food production. As an attempt to test the hypothesis of spatially varying climate change impacts, this study develops a set of climate indices to measure crop yield response across regions. The yield response model based on the climate indices, which are mutually orthogonal, should generate more stable coefficient estimates and yield predictions than models using highly correlated climatic variables. Literature Review Two major methodologies have been used to study the relationship between weather and crop yields: crop growth models and regression models. Crop growth modeling is a computerbased simulation approach based on a mathematical integration of biology, physics, and chemistry (Hoogenboom, 2000; Jones et al., 2003). It incorporates weather information— temperature, precipitation, solar radiation, and humidity—with other factors such as planting and harvest dates, fertilizer and irrigation applications, and soil properties to simulate crop yields. Although useful in examining how weather conditions affect crop growth, crop growth models are typically complex and require extensive, detailed information (Walker, 1989), which makes them less applicable in studies with large spatial scales. Compared with crop growth models, regression models have fewer data demands (Horie, Yajima, and Nakagawa, 1992; Kandiannan et al., 2002; Tannura, Irwin, and Good, 2008). Nonetheless, developing a multiple regression model requires determining the appropriate set of weather factors affecting crop yields. Previous literature has identified the importance of temperature and precipitation (Lobell, Cahill, and Field, 2007) and their nonlinear effects on crop yields (Schlenker and Roberts, 2009). For example, although water is necessary for plant growth, excessive precipitation events can dramatically reduce crop production (Rosenzweig et al., 2002). High temperatures reduce soil moisture, which negatively impacts crop yields, but these impacts may be offset by either precipitation or supplemental irrigation (Mitchell et al., 1990). To allow for nonlinear effects of temperature and precipitation, these variables are often modeled as quadratic forms (Tannura, Irwin, and Good, 2008). Additionally, extreme weather events are likely to reduce crop yields (Porter and Semenov, 2005), which can be partially addressed by using differences between mean daily maximum and minimum temperatures. An issue of regression model specification is the large number of possible independent variables, which consumes degrees of freedom. For example, using monthly data to model a 7-month growing season will result in 14 linear monthly temperature and precipitation variables in the model. This may result in unstable estimates if the sample size is small. The issue of having too many independent variables becomes more severe when quadratic and different terms for weather variables are introduced into the model. Some studies reduce the number of weather variables by using growing degree-days (GDD). The appropriate method for calculating GDD for a given crop, however, is still under debate (Schlenker and Roberts, 2009). Ultimately, a model based on GDD may lead to better Cai et al.: Crop Yield Response to Climate estimates of yield changes, but that is not necessarily the case; two of the GDD-based models estimated by Schlenker and Roberts (2009), for example, did not perform as well as a host of other model specifications from a root mean square error perspective. However, three other model specifications grounded in the concept of GDD performed the best among all the models they examined. Alternatively, statistical variable selection methods are used to reduce the number of variables (Kaufmann and Snell, 1997; Tannura, Irwin, and Good, 2008). A disadvantage of statistical variable selection methods is that they exclusively lean on data while ignoring the agronomic implications of different months within a growing season; it is possible to inadvertently drop agronomically important months. Crop growth is a cumulative process, and weather conditions in any growing month affect crop yields, an argument for retaining all the growing months in the model. Furthermore, weather variables are usually highly correlated; applying statistical variable selection methods to a model with severe multicollinearity could generate unstable estimates. In this article, we construct a set of climate indices using principal component analysis (PCA). Each of these climate indices is a linear combination of all the weather variables (Jolliffe, 2002, p. 169) and thereby retains the influence of all the growing months. Although there is always the potential for omitted variable bias in a regression model, because the PCA model consists exclusively of weather variables, it is unlikely to suffer this fate—any omitted variable is unlikely to be correlated with the climate indices. Furthermore, because the indices generated by PCA are mutually orthogonal, issues related to multicollinearity are avoided (Jolliffe, 2002, p. 167). PCA ranks climate indices according to the magnitude of their variances. To reduce the number of indices, many researchers keep the first several indices with large variances in their models (Baigorria et al., 2008; Gurmessa and Bardossy, 2009; Martinez, Baigorria, and Jones, 2009). However, the indices with larger variances are not necessarily more important than the indices with smaller variances in the regression 721 models. Hadi and Ling (1998) demonstrated that it is possible for the index with the smallest variance to be the only index correlated with a response variable. Therefore, regardless of the ranking of the indices, we use statistical variable selection methods to select the climate indices (Jolliffe, 2002, p. 177).2 Here again PCA avoids one of the traps of statistical variable selection methods; because the PCA-generated indices are orthogonal, dropping one or a set of indices will not bias the parameter estimates of those remaining in the model. As the first attempt of applying PCA to a weather-crop yield model, Cornia and Pochop (1975) used PCA to predict Wyoming winter wheat yields. Their model includes 31 weatherbased principal components and accounts for 54% of the variation in yield data across eight counties. More recently, Kantanantha, Serban, and Griffin (2010) used PCA to study the linkage between weather and crop yields. In their model, temperature and precipitation variables were independently considered for climate indices. We generate climate indices by using both temperature and precipitation variables to ensure that multicollinearity problems are avoided. Our model is also different in that it considers possible nonlinear relationships between crop yields and weather. Methodology Climate Index For each county, a principal component regression (PCR) model is developed to study the response of crop yield to weather variations, where climate indices constructed by PCA are 2 Jolliffe (2002) discusses several decision rules for choosing a subset of principal components in pages 173–177. However, Jolliffe (2002) also mentions on page 177 that ‘‘It is difficult to give any general advice regarding the choice of a decision rule for determining M (the number of explanatory variables). It is clearly inadvisable to base the decision entirely on the size of variance; conversely, inclusion of highly predictive PCs can also be dangerous if they also have very small variances.’’ 722 Journal of Agricultural and Applied Economics, November 2013 used instead of original weather variables.3 Equation (1) is a conventional regression model using original weather variables as predictors of detrended yield, y: (1) y 5 Xb 1 e, variable selection, a climate index is retained in the model if it is significant at the 10% level (Jolliffe, 2002, p. 177). A reduced PCR model is: (4) where X is a matrix of p weather variables with n observations; y is the detrended crop yield; b is a vector of p regression coefficients; and is a vector of error terms. The weather variables in the X matrix include monthly mean temperature, a square term of monthly mean temperature, total monthly precipitation, a square term of total monthly precipitation, and the difference between monthly mean maximum and minimum temperatures for the growing season. Transforming Equation (1) into a PCR model yields: y 5 Zk g k 1 ek , |{z} |{z} |{z} ðn1Þ ðnkÞ ðk1Þ where Zk are the selected climate indices, an ðn kÞ matrix; g k is a coefficient vector of k elements associated with the indices; and k is the error term. Forecasting To forecast the effects of climate change on crop yields, three climate projections, Fi , are appended to the historical data X to generate three sets of combined data Ci : 2 0 y 5 Xb 1 e 5 XAA b 1 e |{z} (2) ðn1Þ X |{z} A 5 ½Xa1 , Xa2 , . . . , Xap , Z 5 |{z} |{z} ðnpÞ 3 A pooled spatial regression framework conducted at the state level is an alternative approach that would explicitly account for potential spatial dependency. In this article, each county is modeled individually, allowing us to take advantage of county-level climate projections. Although a county’s yield is likely to be correlated with the yield of neighboring counties, those yields do not affect each other. That is, yield across counties is not ‘‘spatially dependent.’’ Likewise, because our explanatory variables are unique principal components for each county, the explanatory variables across counties are independent of each other. As a result, potential spatial dependence would occur through the disturbances. If present, our parameter estimates would no longer be efficient but would still be unbiased and consistent (Elhorst, 2012). We recognize that not using a pooled regression framework is a potential limitation of this study. 3 6 ðnpÞ 7 7 56 4 Fi 5, |{z} i 5 1, 2, 3. ðmpÞ The Ci s are then transformed into climate indices Z i by PCA: ðnpÞ ð ppÞ where A is a matrix of eigenvectors of the correlation matrix of X; Z is a matrix whose columns are climate indices, which is used instead of X for model estimation. g is the coefficient vector for Z. Z has the same dimension (n pÞ as X. Based on a stepwise Ci |{z} ððn1mÞpÞ g 1 e, 5 |{z} Z |{z} ðnpÞ ð p1Þ (3) (5) X |{z} Zi |{z} 5 ððn1mÞpÞ (6) Ci |{z} Ai |{z} ððn1mÞpÞ ðppÞ 5 Ci ai1 , Ci ai2 , . . . , Ci aip 2 _ 3 Zi 6 ðnpÞ 7 54 5, i 5 1, 2, 3. €i Z ðmpÞ where Z_ i represents the part of climate indices generated based on historical weather data and Z€i represents the part of climate indices generated based on climate projections. Equation (7) represents the final PCR model: (7) _ i gi 1 e, y 5 Z |{z} |{z} |{z} ðn1Þ i 5 1, 2, 3. ðnpÞ ð p1Þ where each climate projection has its own index matrix Z_ i with dimension ðn ki Þ. ki is the number of indices left in the model after variable selection. After the estimation, future crop yields are projected using Z€i . It should be noted that each county has its own PCR model. It should be noted that the climate indices are constructed by applying PCA to a combined Cai et al.: Crop Yield Response to Climate 723 data set of historical and future climate (Equation [5]). Thus, although only one set of historical weather data exists, the historical parts of these three sets of principal components are different from one another (Equation [6]). This approach avoids applying eigenvectors based exclusively on historical data to future data. Table 1 compares the predictive performance of our approach (Equation [7]) with that of applying the eigenvectors of historical data to future data (Equation [4]). We use the 1960–1999 observations to estimate the PCR model and use the 2000–2009 observations to compare the predictive performance for our approach (Equation [7]) and the approach of applying eigenvectors based on historical data to future data (Equation [4]). The models based on Equation (7) outperform those of Equation (4) in every state for both crops based on mean squared error. This may be the result of the fact that Equation (7) standardizes the data using the mean and standard deviation across all data points. Equation (4), in contrast, standardizes the historical data using the historical mean and standard deviation and uses those moments to convert the future data. As a result, the converted future data of Equation (4) are not actually standardized, i.e., they are not restricted to having a mean of zero and a standard deviation of one. As a result, we use the model described in Equation (7) for the analysis that follows. The projected crop yields from Equation (7) are used to generate a Climate Change Impact Index (CCII). Forty-one years (2010–2050) of projected crop yields are compared with the historical average. The number of years for which future climate scenarios generate lower crop yields as compared with the historical average is recorded for each county. The countylevel CCII is generated by dividing the number of these particular years by the total number Table 1. The State Average of Mean Squared Errors of Two Alternative Principal Component Regression Approaches Based on Equations (4) and (7) Corn Northern states Illinois Indiana Iowa Minnesota Nebraska Southern states Alabama Arkansas Georgia Louisiana Mississippi North Carolina South Carolina Texas Tennessee Soybeans Equation (4) Equation (7) Equation (4) Equation (7) 6,658 82,497 22,519 6,668 5,209 712 581 653 804 434 22,168 12,110 620 1,942 763 68 54 46 78 51 40,306 25,178 4,700 50,418 2,977 9,185 15,279 5,338 2,443 848 641 854 837 560 884 762 797 623 1,824 428 408 733 2,880 947 814 10,113 235 113 38 107 53 60 56 48 116 84 Note: In Equation (7), the climate indices are constructed by applying principal component analysis (PCA) to a combined data set of historical and future climate. In Equation (4), the climate indices are constructed by applying PCA only to historical data. The data from 1960 to 2009 were broken up into two sets: a set of ‘‘historical’’ data from 1960 to 1999 and a set of ‘‘future’’ data from 2000 to 2009. The data from 2000–2009 were used to compare the difference between observed yield and predicted yield. The sum of squared errors between observed yield and predicted yield are reported in the table for the two approaches. 724 Journal of Agricultural and Applied Economics, November 2013 Table 2. Leave-One-Out Cross-Validation Results by States and by Crops Corn Soybeans Corn Northern states Illinois Indiana Iowa Minnesota Nebraska Southern states Alabama Arkansas Georgia Louisiana Mississippi North Carolina South Carolina Tennessee Texas 54.8% 70.2% 30.5% 30.3% 41.3% 48.6% 41.7% 52.5% 33.3% 67.6% 48.3% 30.4% 38.9% 45.5% 52.4% 53.6% 81.8% 64.6% 32.0% 61.5% 36.4% 44.7% 26.7% 55.6% 44.9% 17.9% 47.6% 47.4% Note: The county-level observed yields were regressed on predicted yield generated by the leave-one-out method. The numbers in the table show the proportion of counties with significant relationships at significance level of 0.1 within each state. of future years. Equation (8) represents the state-level, area-weighted CCII: Xcs (8) CCII s 5 Table 3. The Average R2 Values of Principal Component Regression Models by States and by Crops Based on Equation (7) ui i51 41 Xcs i51 * Ai Ai , where s denotes specific U.S. states; cs denotes number of counties in specific states; u denotes the number of years for which a certain climate scenario generates lower crop yields as compared with the historical average; and A denotes the county-level harvested acreage. A county with higher harvested acreage is given more weight. A CCII higher than 0.5 indicates that future climate will have a net negative effect on crop yields, whereas a CCII lower than 0.5 indicates that future climate will have a net positive effect on crop yields. Data We use county-level crop yields and weather data covering 50 growing seasons (1960–2009). Northern states Illinois Indiana Iowa Minnesota Nebraska Southern states Alabama Arkansas Georgia Louisiana Mississippi North Carolina South Carolina Texas Tennessee Soybeans R2 Adjusted R2 R2 Adjusted R2 0.76 0.72 0.63 0.60 0.63 0.69 0.65 0.55 0.52 0.56 0.73 0.66 0.50 0.61 0.67 0.68 0.59 0.43 0.53 0.60 0.75 0.46 0.69 0.57 0.55 0.70 0.71 0.67 0.69 0.70 0.39 0.63 0.51 0.48 0.64 0.65 0.60 0.63 0.75 0.63 0.76 0.63 0.69 0.77 0.69 0.62 0.74 0.69 0.56 0.73 0.56 0.62 0.72 0.63 0.55 0.68 Major U.S. producing states for corn and soybeans are selected consisting of five northern states. Major producing states in the south are also selected to test the hypothesis of spatially varying yield responses to future climatic changes.4 The historical weather data for each growing season, including monthly mean temperature, monthly mean minimum temperature, 4 Northern states include Illinois, Indiana, Iowa, Minnesota, and Nebraska; southern states include Alabama, Arkansas, Georgia, Louisiana, Mississippi, North Carolina, South Carolina, Tennessee, and Texas. The selection of the states is based on production criterion for corn and soybeans. Illinois, Indiana, Iowa, Minnesota, and Nebraska are the top five soybeanproducing states; their total production exceeds 50% of U.S. total production for 2007–2009. Illinois, Iowa, Minnesota, and Nebraska are the top four corn-producing states; their total production exceeds 50% of U.S. total production for 2007–2009. Indiana was included in the corn analysis for consistency. Because we focus on a north–south yield response comparison, we use nine contiguous cornand soybean-producing states across the southern region. Cai et al.: Crop Yield Response to Climate 725 Table 4. Climate Change Impact Index by States and by Crops Corn Northern states Illinois Indiana Iowa Minnesota Nebraska North averages Southern states Alabama Arkansas Georgia Louisiana Mississippi North Carolina South Carolina Tennessee Texas South averages Soybeans CSIRO 3.5 MIROC 3.2 CSIRO 3.5 MIROC 3.2 0.569 0.610 0.501 0.430 0.542 0.530 0.659 0.776 0.529 0.468 0.577 0.602 0.584 0.473 0.461 0.466 0.519 0.501 0.587 0.523 0.422 0.365 0.599 0.499 0.517 0.565 0.575 0.652 0.520 0.491 0.552 0.576 0.604 0.561 0.801 0.534 0.701 0.670 0.618 0.782 0.866 0.787 0.676 0.715 0.703 0.623 0.609 0.746 0.687 0.599 0.621 0.632 0.580 0.644 0.832 0.732 0.738 0.761 0.796 0.796 0.720 0.789 0.588 0.750 Note: The numbers in this table represent the proportions for which certain climate scenario generates lower crop yield as compared with the historical average at the state level. monthly mean maximum temperature, and monthly total precipitation, were retrieved from the NOAA National Climate Data Center (2011). Climate projections were developed by the USDA Forest Service as part of the 2010 Renewable Resources Planning Act assessment of U.S. natural resource demand and supply. These climate projections, covering the period 2001–2100, were derived from three GCMs: CGCM 3.1, CSIRO 3.5, and MIROC 3.2, assuming the SRA1B scenario from the Special Report on Emission Scenarios of IPCC (IPCC, 2007; Coulson et al., 2010). Although the climate change projection is available up to year 2100, we limit our time horizon to 2050, recognizing that the reliability of forecast is often inversely related to the time horizon. Annual corn and soybean yield data were retrieved from U.S. Department of Agriculture, National Agricultural Statistical Service (2011) from 1960 to 2009 at the county level. Corn and soybeans were chosen because they are major crops for many northern and southern states. Crop yield data were detrended to a 2009 technology level. 5 Results Model Validation The leave-one-out cross-validation method (Efron and Gong, 1983) is conducted to test the validity of the proposed PCR models. For 5 Technological change improves crop yields over time. Previous models generally include additional predictors to proxy for technology change. Possible candidates for this predictor include gross domestic product and time trends (Choi and Helmberger, 1993; McCarl, Villavicencio, and Wu, 2008). We study the relationship between weather and detrended crop yields. Specifically, time-series crop yields are regressed over polynomial time trends, and the fitted yield trend is adjusted to the 2009 level. yieldtrend 5 b0 1 tb1 1 t2 b2 1 e, t 5 1, . . . . . . , 50. b 1 yield yielddetrended 5 yield yield trend 2009 726 Journal of Agricultural and Applied Economics, November 2013 Table 5. Historical and Projected Corn Yields by State Corn 50-year Historical CSIRO 3.5 CGCM 3.1 MIROC 3.2 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 186.19A 41 179.55B –3.57% 11.93 168.88 188.96 41 177.26B –4.80% 6.30 172.73 181.51 41 173.87B –6.62% 7.17 168.76 179.47 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 170.15A 41 161.41B –5.14% 12.83 151.84 172.28 41 159.79BC –6.09% 7.44 153.94 165.35 41 153.57C –9.74% 9.93 147.37 163.21 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 188.47A 41 187.18A –0.68% 7.47 183.15 192.17 41 184.71A –2.00% 6.25 182.26 188.42 41 185.06A –1.81% 4.72 183.14 187.56 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 179.11A 41 185.14B 3.37% 5.91 181.60 189.76 41 183.05AB 2.20% 6.25 179.28 188.87 41 182.69AB 2.00% 4.79 179.89 186.42 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 174.05A 41 171.63A –1.39% 7.13 166.45 176.90 41 172.55A –0.86% 4.87 168.78 176.26 41 170.72A –1.91% 4.54 168.13 173.50 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 99.18A 41 95.66AB –3.55% 12.30 88.29 101.06 41 89.84B –9.42% 13.77 77.31 101.14 41 77.14C –22.22% 13.03 69.05 83.54 41 151.54A 0.37% 8.31 146.45 155.74 41 153.44A 0.87% 4.12 152.04 156.30 41 153.01A 0.59% 3.40 150.88 155.28 Illinois 15.08 179.71 196.64 Indiana 13.68 14.49 178.92 Iowa 13.39 183.07 197.59 Minnesota 15.11 175.09 189.16 Nebraska 10.71 167.69 183.21 Alabama 14.75 90.11 106.31 Arkansas N Meana Percentage change Standard deviation Lower quintile Upper quintile 49 152.11A 10.90 147.22 158.58 Cai et al.: Crop Yield Response to Climate 727 Table 5. Continued Corn 50-year Historical CSIRO 3.5 CGCM 3.1 MIROC 3.2 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 132.79A 41 125.82AB –5.25% 9.20 121.35 133.04 41 121.14B –8.77% 8.98 114.08 129.95 41 114.47C –13.80% 8.08 108.84 119.37 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 141.06A 41 134.00B –5.00% 8.84 128.26 139.83 41 139.60A –1.04% 7.47 133.88 142.14 41 132.99B –5.72% 6.58 129.16 135.37 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 110.47A 41 112.18A 1.55% 6.63 108.83 115.41 41 110.72A 0.23% 5.85 107.60 113.76 41 105.58B –4.43% 4.83 101.43 109.33 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 111.17A 41 105.75AB –4.88% 14.22 95.37 115.39 41 101.09B –9.07% 8.98 94.42 107.31 41 89.14C –19.82% 9.52 84.27 92.57 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 101.73A 41 96.70AB –4.94% 20.60 88.64 109.66 41 89.90B –11.63% 14.36 77.62 100.30 41 76.84C –24.47% 15.50 68.03 85.65 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 128.81A 41 122.21B –5.12% 9.72 114.34 129.07 41 120.11B –6.75% 11.38 110.73 126.39 41 110.85C –13.94% 11.37 104.84 119.39 N Meana Percentage change Standard deviation Lower quintile Upper quintile 42 161.97A 41 158.70A –2.02% 7.43 153.60 163.88 41 157.39AB –2.83% 11.42 148.69 165.81 41 151.51B –6.46% 6.49 147.65 154.98 Georgia 12.14 126.53 142.10 Louisiana 13.53 131.72 149.92 Mississippi 10.13 104.53 116.37 North Carolina 14.38 99.82 120.84 South Carolina 16.87 92.30 115.15 Tennessee 12.63 121.29 140.21 Texas 17.59 149.47 175.15 a The Tukey–Kramer testing method is used here. The same letter indicates that the means are not significantly different from each other at a significance level of 0.05. 728 Journal of Agricultural and Applied Economics, November 2013 Table 6. Historical and Projected Soybeans Yields by State Soybeans 50-year Historical CSIRO 3.5 CGCM 3.1 MIROC 3.2 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 48.92A 41 47.16B –3.60% 2.23 45.43 48.75 41 47.93AB –2.02% 1.57 46.91 49.32 41 46.15C –5.66% 2.03 45.12 47.45 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 49.52A 41 49.67A 0.30% 2.05 47.75 51.12 41 49.35A –0.34% 1.69 47.83 50.72 41 48.97A –1.11% 1.73 47.86 49.96 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 51.00A 41 51.30A 0.59% 1.69 50.26 52.26 41 51.59A 0.12% 1.52 50.51 52.72 41 51.91A 1.78% 1.49 50.86 52.86 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 44.95A 41 45.75AB 1.78% 1.83 44.84 47.19 41 46.92BC 4.38% 1.66 46.22 48.05 41 47.88C 6.52% 1.54 46.96 48.53 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 49.68A 41 49.98A 0.60% 3.39 47.73 52.47 41 49.42A –0.52% 2.87 47.75 51.41 41 48.42A –2.54% 2.46 46.37 50.64 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 32.84A 41 27.60B –15.96% 4.28 24.24 30.71 41 26.48B –19.37% 3.97 24.34 30.07 41 22.81C –30.54% 5.12 18.61 25.93 N Meana Percentage change Standard deviation Lower quintile Upper quintile 49 38.46A 41 37.45AB –2.63% 1.87 36.45 38.40 41 36.72BC –4.52% 2.92 35.10 38.69 41 35.40C –7.96% 1.75 34.06 36.81 Illinois 3.52 47.83 51.25 Indiana 3.63 47.08 51.80 Iowa 3.85 50.10 53.36 Minnesota 4.27 43.25 47.42 Nebraska 4.23 46.28 52.84 Alabama 4.93 30.12 36.21 Arkansas 2.99 36.81 40.13 Cai et al.: Crop Yield Response to Climate 729 Table 6. Continued Soybeans 50-year Historical CSIRO 3.5 CGCM 3.1 MIROC 3.2 N Meana Standard deviation Percentage change Lower quintile Upper quintile 50 30.75A 4.14 41 28.14B 2.90 –8.49% 26.49 30.41 41 26.97B 3.30 –12.29% 24.37 30.11 41 24.25C 3.02 –21.14% 22.07 26.05 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 38.58A 41 34.48B –10.63% 3.47 31.68 36.33 41 35.67B –7.54% 1.91 34.42 37.26 41 34.47B –10.65% 2.17 32.89 35.48 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 38.95A 41 36.52B –6.24% 3.38 34.25 38.28 41 35.51B –8.83% 3.80 31.98 38.42 41 32.33C –17.00% 3.35 29.70 34.58 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 30.04A 41 27.01B –10.09% 3.26 25.18 29.60 41 27.97B –6.89% 1.95 27.17 29.42 41 23.97C –20.21% 2.52 22.43 25.61 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 26.14A 41 24.58B –5.97% 2.68 22.81 26.47 41 24.32B –6.96% 2.42 23.11 25.82 41 21.06C –19.43% 2.20 19.70 22.23 N Meana Percentage change Standard deviation Lower quintile Upper quintile 42 26.29A 41 25.78AB –1.94% 1.50 24.67 26.78 41 25.97A –1.22% 1.89 24.85 27.35 41 24.32B –7.49% 1.93 23.21 25.33 N Meana Percentage change Standard deviation Lower quintile Upper quintile 50 37.02A 41 33.56B –9.35% 4.19 30.42 36.06 41 32.28BC –12.80% 4.75 29.67 34.68 41 30.23C –18.34% 3.88 27.37 33.31 Georgia 27.70 34.27 Louisiana 3.25 36.56 40.91 Mississippi 4.04 36.24 41.79 North Carolina 2.85 28.61 32.22 South Carolina 3.28 24.10 28.53 Texas 3.87 23.24 29.42 Tennessee 4.71 35.33 39.29 a The Tukey–Kramer testing method is used here. The same letter indicates that the means are not significantly different from each other at a significance level of 0.05. 730 Figure 1. Journal of Agricultural and Applied Economics, November 2013 Histograms of Annual Average Historical and Future Corn Yields for Northern Counties each county, one year is removed from the sample and the model is estimated using data from the remaining years. Then the climate indices for the year removed are entered into the model to predict the crop yield for that year. This was repeated for each historical year. For each county, a regression of observed yields on predicted yields is conducted. Then the proportion of counties in each state with regressions significant at the 0.1 level was recorded. The results of cross-validation indicate that approximately half of the counties show significant relationships between observed and predicted yields (Table 2). This result is acceptable given that only a part of crop yield variation is weatherrelated. The percentage of counties with a significant relationship between observed and predicted yields varies by state and by crop (e.g., 30% compared with 70% for corn in Minnesota and Indiana, respectively). Two possible reasons for this variation are: different states have different levels of agricultural technology and environmental conditions, which determine how much weather affects crop yields; and weather data may be collected from stations located at a considerable distance from some of the production area in a county and therefore could miss accurate climate conditions.6 Besides crossvalidation, we also observe that R2 values of the PCR models range from 0.46 to 0.77 at the 6 Preferred climate data for our model would be average weather measured only at the farms in a county (Data 1). Two other alternatives are the county average of gridded weather data (Data 2) and station-level weather data (Data 3). Both Data 2 and 3 are unbiased estimates of Data 1, assuming both farms and weather stations are randomly located in the county. Cai et al.: Crop Yield Response to Climate Figure 2. 731 Histograms of Annual Average Historical and Future Corn Yields for Southern Counties state level, showing the overall goodness of fit for the regression models (Table 3). Forecasts The crop yields are projected by three GCMs: CSIRO 3.5, CGCM 3.1, and MIROC 3.2 under the SRA1B scenario. Most states have a CCII larger than 0.5, indicating negative effects of future climate on crop yields (Table 4). MIROC 3.2 (the warmest scenario) generates higher CCII than CSIRO 3.5 (the coldest scenario) in most states. Northern states generally have a lower CCII value than southern states, indicating that global warming is potentially a more severe problem for corn and soybean yields in low latitude regions. This north–south difference is larger for the warmer climate scenario. We also observe that some CCII are less than 0.5, indicating positive effects of future climate on crop yields, generally in northern states. In those southern states with small CCII such as Arkansas and Mississippi, we notice that they have soil types (Alfisols) that are more similar to the northern states as compared with rest of the southern states (U.S. Department of Agriculture, Natural Resources Conservation Service, 2012). These results suggest the effects of climate change on corn and soybean yields will be spatially heterogeneous and that soil type may be an important indicator of the magnitude of yield effects. However, it is important to note that the identification approach pursued here prevents us from formally testing the statistical significance of this finding. In addition to CCII in which the effects of climate change are indicated by the proportions 732 Figure 3. Journal of Agricultural and Applied Economics, November 2013 Histograms of Annual Average Historical and Future Soybean Yields for Northern Counties that predicted crop yields are higher or lower than the historical average, we generate the mean yield percentage changes between predicted crop yields and the historical average (Tables 5 and 6; Figures 1–4). In general, warmer scenarios induce larger yield reductions than cooler scenarios, and southern states have larger yield reductions than northern states. Again, yield reductions in Arkansas and Mississippi are similar to those in northern states, which may be explained by similar soil types. Minnesota is the only state where both corn and soybean yields increase in the climate scenarios. Minnesota has relatively lower historical temperatures compared with other states; thus, future global warming could move the temperature toward its optimal value for crop growth. This result is consistent with findings from Almaraz et al. (2008). They found that the optimum temperatures for corn yield are approximately 1°C above normal for the Montérégie region of southwestern Quebec, Canada, which has comparable latitude as Minnesota.7 The same justification can be used for Iowa and Nebraska, which have the second and third highest latitudes, respectively, out of the 14 states and have no significant changes of corn and soybeans yields. Although Georgia and Alabama are neighboring states located at similar latitudes, Alabama’s crop yields are projected to experience worse losses. For example, corn (soybean) 7 The latitude of Montérégie is approximately 45°N, whereas the latitude of Minnesota ranges from 43° 309 N to 49° 239 N. Cai et al.: Crop Yield Response to Climate Figure 4. 733 Histograms of Annual Average Historical and Future Soybean Yields for Southern Counties yield loss for Georgia is 13.80% (21.14%) compared with a loss of 22.22% (30.54%) for Alabama’s corn (soybean). The worse future crop yields projection for Alabama compared with Georgia could be explained by a better agricultural infrastructure in Georgia such as irrigation, which helps maintain crop yields under unfavorable weather conditions.8 However, as with the results reported in Table 4, we do not formally test whether the observed differences are statistically significant. 8 During the census years 1997, 2002, and 2007, only 6.8% (1.4%) of harvested corn (soybeans) was irrigated in Alabama, whereas 37.9% (10.7%) of harvested corn (soybeans) was irrigated in Georgia (Quick Stats, USDA/NASS, accessed October 20, 2011). Crop yields projected by warmer climate projections are not always lower than cooler climate projections, even for southern states such as corn for Louisiana. One explanation could be that temperatures from the warmer projection are not always higher than the cooler projection. For example, from 2010 to 2050 in Louisiana, the coldest projection has a higher temperature than the warmest projection in five years, and it has a higher temperature than the middle projection in 17 years (Figure 5). Similar patterns could also be observed in other states. Another explanation is the complexity of crop yield response modeling resulting from the growth process of crops; therefore, it is hard to implement a comprehensive model that considers all the influential factors. Our model attempts to consider only the part of crop yield variations 734 Journal of Agricultural and Applied Economics, November 2013 Figure 5. Historical Temperature and Future Temperature Projections by Three Global Climate Models (GCMs) under the SRA1B Scenario for Louisiana a spatially varying crop yield–climate change relationship? To answer this question, we simulate crop yields assuming that each state has the same magnitude of climate change. We induced by weather and uses polynomial time trends to control for technological improvements. Are the state differences in yield variations induced by spatially varying climate change or Table 7. Absolute Future Temperature and Precipitation Change from the Historical Average Temperature (Fahrenheit) State Northern states Indiana Illinois Iowa Minnesota Nebraska Southern states Alabama Arkansas Georgia Louisiana Mississippi North Carolina South Carolina Tennessee Texas Precipitation (inches) CGCM 3.1 CSIRO 3.5 MIROC 3.2 CGCM 3.1 CSIRO 3.5 MIROC 3.2 3.32 2.25 3.61 3.92 3.11 2.82 2.11 2.85 3.48 2.42 4.61 2.86 5.19 5.06 4.93 0.00 0.27 0.07 0.08 0.04 0.00 0.47 0.08 0.08 0.20 –0.26 0.15 –0.26 –0.05 –0.24 2.27 2.91 2.45 2.10 2.55 2.42 2.46 2.50 3.46 1.99 2.57 2.10 2.04 2.36 2.05 2.12 2.08 2.92 3.50 4.92 3.81 3.39 3.94 3.48 3.78 3.93 5.05 0.13 0.12 0.09 0.11 0.15 0.19 0.19 0.26 0.08 0.21 0.14 0.16 0.09 0.18 0.16 0.27 0.22 –0.04 –0.59 –0.39 –0.64 –0.63 –0.47 –0.29 –0.40 –0.16 –0.25 Note: Historical average is based on 1960–2009. Future average is based on 2010–2050. Cai et al.: Crop Yield Response to Climate 735 Table 8. Percentage Change of Corn and Soybeans Yields under the Same Climate Change across States Northern states Illinois Indiana Iowa Minnesota Nebraska North averages Southern states Alabama Arkansas Georgia Louisiana Mississippi North Carolina South Carolina Tennessee Texas South averages Corn Soybeans –1.5% –1.3% –0.2% 0.5% –0.6% –0.6% –1.3% –0.4% 0.6% 0.8% –0.1% –0.1% –4.4% –0.5% –3.4% –1.5% –1.8% –3.2% –7.8% –3.1% –2.8% –3.2% –4.8% –1.9% –2.2% –4.3% –3.4% –2.7% –1.7% –4.2% –0.9% –2.9% Note: Temperature changes are random numbers from a normal distribution with a mean of 3.192°F and a standard deviation of 1.539, and precipitation changes are random numbers from a normal distribution with a mean of 0.035 inches and a standard deviation of 0.615 for all the states. For each state, 100 random numbers are withdrawn to simulate crop yields. The numbers reported in the table are the average of simulated yield percentage change. assume a 3.192°F increase of temperature with a standard deviation of 1.539 and a 0.035-inch decrease of precipitation with a standard deviation of 0.615 for all the states. These values are the average future temperature and precipitation changes obtained from three climate scenarios for 14 states (Table 7). We observe that the crop yield reductions are spatially varying even under the same magnitude of climate change based on 100 simulations (Table 8). We find that all the states but Minnesota have reduced corn and soybean yields. Southern states generally have larger crop yield reductions than northern states, whereas Arkansas, Mississippi, and Louisiana’s yield responses tend to be similar to the northern states. Given that actual climate scenarios vary across the states, we conclude that both spatially varying climate change and a spatially varying crop yield– climate change relationship contribute to the state differences in crop yield reductions under climate scenarios. Conclusions To forecast crop yields under future climate scenarios, county-specific PCR models were estimated in 14 U.S. states. Crop yields were forecasted in response to three GCMs: CSIRO 3.5 (the coldest), CGCM 3.1, and MIROC 3.2 (the warmest). Our results indicate: 1) future climate scenarios generally have modest effects on crop yields in the northern states while negatively affecting crop yields in the southern states; 2) warmer climate scenarios generate lower crop yields; 3) the north–south differences in climate change effects are larger for warmer scenarios; and 4) soil type may explain why some southern states have modest yield responses across climate scenarios. We make some major assumptions. It is assumed that there is no CO2 fertilization effect for crop growth. Some actual field research indicates a small increase in crop yields under higher CO2 concentration (Long et al., 2006). We exclude CO2 fertilization effect not only to simplify the model, but also as a result of the inability of specifying the magnitude of this effect. Schlenker and Roberts (2009) stated that a CO2 effect might be part of the time trend, which is statistically entangled with technological change. Therefore, they also do not model a CO2 effect. We also realize that total precipitation could only partially capture the extreme precipitation events. Besides the total precipitation, the timing of precipitation is also expected to be important. As a result of the availability of climate variables in climate projections we used, we did not fully capture the effects of drought or flood in the model. We also do not account for potential spatial dependence across counties. Although this is unlikely to affect the unbiasedness and consistency of our estimates, if such dependence is present, they would no longer be efficient. This research contributes to the literature in a number of ways. First, it is one of the first applications of PCA to estimate the linkage between weather and crop yields. We also improve on previous PCR models by including 736 Journal of Agricultural and Applied Economics, November 2013 quadratic terms of weather variables. To the best of our knowledge, these quadratic terms have not been included in previous PCR models. We also use a new method to apply PCR model in a predictive framework, which has higher predictive performance than previous PCR models. Most importantly, we contribute to the literature by demonstrating the potential for spatially varying impacts of climate change on crop yields in several northern and southern U.S. states using county-specific PCR models. These results suggest that returns to research and extension efforts aimed at mitigating the agricultural effects of climate change are likely to differ across regions of the United States. However, we stress that these findings are merely suggestive because we do not formally test for patterns of spatial heterogeneity. [Received April 2012; Accepted April 2013.] References Almaraz, J., F. Mabood, X. Zhou, E. Gregorich, and D. Smith. ‘‘Climate Change, Weather Variability and Corn Yield at a Higher Latitude Locale: Southwestern Quebec.’’ Climatic Change 88(2008):187–97. Baigorria, G.A., J.W. Hansen, N. Ward, J.W. Jones, and J.J. O’Brien. ‘‘Assessing Predictability of Cotton Yields in the Southeastern United States Based on Regional Atmospheric Circulation and Surface Temperatures.’’ Journal of Applied Meteorology and Climatology 47(2008):76–91. Butterworth, M.H., M. Semenov, A.P. Barnes, D. Moran, J.S. West, and B.D.L. Fitt. ‘‘North– South Divide: Contrasting Impacts of Climate Change on Crop Yields in Scotland and England.’’ Journal of the Royal Society, Interface 42(2009):123–30. Choi, J.S., and P.G. Helmberger. ‘‘How Sensitive Are Crop Yields to Price Changes and farm Programs?’’ Journal of Agricultural and Applied Economics 25(1993):237–44. Cornia, R.L., and L.O. Pochop. ‘‘Principal Components and Crop Weather Analysis.’’ Laramie, WY: Wyoming Agricultural Experiment Station, University of Wyoming; 1975. Coulson, D.P., L.A. Joyce, D.T. Price, D.W. McKenney, R. Siltanen, P. Papadopol, and K. Lawrence. Climate Scenarios for the Conterminous United States at the County Spatial Scale Using SRES Scenarios A1B and A2 and Prism Climatology. U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station, Fort Collins, CO, 2010. Internet site: www.fs.fed.us/rm/data_archive/dataaccess/US_ ClimateScenarios_county_A1B_A2_PRISM. shtml (Accessed October 20, 2011). Efron, B., and G. Gong. ‘‘A Leisurely Look at the Bootstrap, the Jackknife, and Cross-Validation.’’ The American Statistician 37(1983):36–48. Elhorst, J.P. ‘‘Dynamic Spatial Panels: Models, Methods, and Inferences.’’ Journal of Geographical Systems 14(2012):5–28. Gurmessa, T.K., and A. Bardossy. ‘‘A Principal Component Regression Approach to Simulate the Bed-Evolution of Reservoirs.’’ Journal of Hydrology (Amsterdam) 368(2009):30–41. Hadi, A.S., and R.F. Ling. ‘‘Some Cautionary Notes on the Use of Principal Components Regression.’’ The American Statistician 52(1998): 15–19. Hoogenboom, G. ‘‘Contribution of Agrometeorology to the Simulation of Crop Production and Its Applications.’’ Agricultural and Forest Meteorology 103(2000):137–57. Horie, T., M. Yajima, and H. Nakagawa. ‘‘Yield Forecasting.’’ Agricultural Systems 40(1992): 211–36. IPCC. ‘‘Climate Change, 2007: Synthesis Report. Contribution of Working Groups I, II and III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change.’’ R.K. Pachauri and A. Reisinger, eds. Cambridge, UK: Cambridge University Press, 2007. Jolliffe, I.T. Principal Component Analysis. 2nd ed. New York, NY: Springer, 2002. Jones, J.W., G. Hoogenboom, C.H. Porter, K.J. Boote, W.D. Batchelor, L.A. Hunt, P.W. Wilkens, U. Singh, A.J. Gijsman, and J.T. Ritchie. ‘‘The DSSAT Cropping System Model.’’ European Journal of Agronomy 18(2003):235–65. Kandiannan, K., K.K. Chandaragiri, N. Sankaran, T.N. Balasubramanian, and C. Kailasam. ‘‘CropWeather Model for Turmeric Yield Forecasting for Coimbatore District, Tamil Nadu, India.’’ Agricultural and Forest Meteorology 112(2002): 133–37. Kantanantha, N., N. Serban, and P. Griffin. ‘‘Yield and Price Forecasting for Stochastic Crop Decision Planning.’’ Journal of Agricultural Biological & Environmental Statistics 15(2010): 362–80. Kaufmann, R.K., and S.E. Snell. ‘‘A Biophysical Model of Corn Yield: Integrating Climatic and Cai et al.: Crop Yield Response to Climate Social Determinants.’’ American Journal of Agricultural Economics 79(1997):178–90. Lobell, D.B., K.N. Cahill, and C.B. Field. ‘‘Historical Effects of Temperature and Precipitation on California Crop Yields.’’ Climatic Change 81(2007):187–203. Long, S.P., E.A. Ainsworth, A.D.B. Leakey, N. Josef, and D.R. Ort. ‘‘Food for Thought: Lower-than-Expected Crop Yield Stimulation with Rising CO2 Concentrations.’’ Science 312(2006):1918–21. Martinez, C.J., G.A. Baigorria, and J.W. Jones. ‘‘Use of Climate Indices to Predict Corn Yields in Southeast USA.’’ International Journal of Climatology 29(2009):1680–91. McCarl, B.A., X. Villavicencio, and X. Wu. ‘‘Climate Change and Future Analysis: Is Stationary Dying?’’ American Journal of Agricultural Economics 90(2008):1241–47. Mitchell, J.F.B., S. Manabe, V. Meleshko, and T. Tokioka. ‘‘Equilibrium Climate Change and its Implications for the Future.’’ The IPCC Scientific Assessment. J.L. Houghton, G.J. Jenkins, and J.J. Ephraums, eds. Cambridge, UK: Cambridge University Press, 1990. NOAA, National Climatic Data Center. Monthly Surface Data (DS3220). Internet site: http:// cdo.ncdc.noaa.gov/pls/plclimprod/somdmain. somdwrapper?datasetabbv5DS3220&country abbv5&georegionabbv5&forceoutside5 (Accessed October 20, 2011). Porter, J.R., and M.A. Semenov. ‘‘Crop Responses to Climatic Variation.’’ Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 360(2005):2021–35. 737 Rosenzweig, C., F.N. Tubiello, R. Goldberg, E. Mills, and J. Bloomfield. ‘‘Increased Crop Damage in the U.S. from Excess Precipitation under Climate Change.’’ Global Environmental Change 12(2002):197–202. Schlenker, W., and M.J. Roberts. ‘‘Nonlinear Temperature Effects Indicate Severe Damages to U.S. Crop Yields under Climate Change.’’ Proceedings of the National Academy of Sciences of the United States of America 106(2009):15594–98. Tannura, M.A., S.H. Irwin, and D.L. Good. ‘‘Weather, Technology, and Corn and Soybean Yields in the U.S. Corn Belt.’’ Marketing and Outlook Res. Rep. No. 2008-01, Dept. of Agr. Econ., University of Illinois at UrbanaChampaign, 2008. Tao, F., M. Yokozawa, Y. Xu, Y. Hayashi, and Z. Zhang. ‘‘Climate Changes and Trends in Phenology and Yields of Field Crops in China, 1981–2000.’’ Agricultural and Forest Meteorology 138(2006):82–92. U.S. Department of Agriculture, National Agricultural Statistics Service. Quick Stats—The National Agricultural Statistics Service Interactive, Online Statistical Database. Internet site: http://quickstats.nass.usda.gov/ (Accessed October 20, 2011). U.S. Department of Agriculture, Natural Resources Conservation Service. Technical References. Internet site: http://soils.usda.gov/technical/ classification/orders/alfisols_map.html (Accessed October 10, 2012). Walker, G.K. ‘‘Model for Operational Forecasting of Western Canada Wheat Yield.’’ Agricultural and Forest Meteorology 44(1989):339–51.